Maybe a little bit of perspective on why prices and volumes are presented as 'adjusted'. Let's take a real life example of AAPL which had a 7-for-1 stock split on June 9, 2014. The real, you would have paid this much, 'un-adjusted' close prices for AAPL around the time of the split were
Date Equity price
2014-06-05 Equity(24 [AAPL]) 644.82
2014-06-06 Equity(24 [AAPL]) 647.35
2014-06-09 Equity(24 [AAPL]) 92.22
2014-06-10 Equity(24 [AAPL]) 93.70
On Friday 2014-06-06 one would have paid $647.35 for a share of AAPL. On Monday 2014-06-09 one would have paid $92.22 for a share. The reality is that on Monday there were 7 times as many shares in circulation so each share was worth proportionately less. Conversely, if one had purchased a share of AAPL on Friday then they would hold 7 shares on Monday. Nonetheless, those are what a share would have cost on those days.
In a 'simple' world that would be the end of the story. That was the price. Period.
However, a huge part of trading revolves around returns and gains. 'How much did I make ' and 'how much is it worth now' are fundamental questions. Looking simply at the prices above it would appear AAPL lost a lot on Monday 2014-06-09. That clearly wasn't the case. To get around this, the concept of 'adjusted prices' was invented. First one needs to specify the 'adjusted as of date'. Price and volumes are typically adjusted backwards from that date. Next one needs to know when any stock splits occurred. Finally, divide the price by the split value if the split (or splits) occurred before a particular date in question. Below are the prices adjusted as of 2014-06-09.
Date Equity price
2014-06-05 Equity(24 [AAPL]) 92.12
2014-06-06 Equity(24 [AAPL]) 92.48
2014-06-09 Equity(24 [AAPL]) 92.22
Now one can see there was actually a relative increase from 2014-06-05 to 2014-06-09. Additionally, the volumes would have also been proportionally adjusted.
Now the case if prices are adjusted as of 2014-06-06.
Date Equity price
2014-06-05 Equity(24 [AAPL]) 644.82
2014-06-06 Equity(24 [AAPL]) 647.35
Since the split hasn't occurred yet the prices appear the same.
A similar calculation is done whenever adjusting for dividends. It looks much like a small stock split. It adjusts share values before the dividend down by the amount of the dividend. It's as if the dividend was a rebate the company paid on it's stock so the net price was a little lower.
OK, back to the original question... "If the pipeline was looking for the daily volume of 'AAPL' for 2017-01-04', how would the adjusting be different in both cases?". Pipeline would use 2017-01-04 as the adjustment date while Yahoo uses the current date (ie 2019-8-22). Pipeline wouldn't include any corporate actions after 2017-01-04 while Yahoo would.
Using the example above, in the research (notebook) environment a pipeline returns a multi-indexed dataframe. All pipeline data based upon price or volume are adjusted daily as of the pipeline date (ie the date in the level 0 index). Using the example above, for index 2014-06-06 the latest price would be adjusted as of that date and be 647.35. For index 2014-06-09 the latest price would be adjusted as of that date and be 92.22. The 2014-06-06 price would be similarly adjusted and be 92.48. Think of it as a 'rolling' adjustment as-of date. To prevent lookahead bias, the entire framework is based around having pipeline return only data that would have been known on that date. It never looks into the future to assume any events.
To get an 'apples to apples' comparison (please excuse the pun) use the Quantopian get_pricing method with and end date equal to the current date. This should closely match Yahoo or other 3rd party data since they will all be adjusted to the same current date.
Now, there may be other reasons why volumes vary so these need to be looked at case by case. As mentioned, if that doesn't seem to be the issue you are seeing please attach a notebook showing the problem.
A final word of caution on Yahoo data. Since it's constantly being re-adjusted as of the current date, historical data downloaded six months ago may be different than what is downloaded today. It all depends upon if any relevant corporate actions occurred in the past 6 months. This is one aspect of Quantopian data which is important in research. It can be replicated by specifying the as-of date.
